|
|
|
This page presents some of the products used in photo-enforcement. Material is provided by vendors. The page is divided into four section broadly grouping the products. Some products overlap the group they are in. For information about any specific product on this page, please contact that vendor. Page Contents:
MegaPixel is unique in that it provides solutions based neuro-like algorithms that function in a manner similar to human optic system. License Plate Recognition System MegaPixel has been developing license plate recognition systems since 1995. The systems were first developed for recognition of Russian plates which are often more difficult to read that those in other countries. Russian license plates exist in 16 combinations of digits and letters, which can be both large and small, black and white:
New license plate types
Transit license plate types Plates for countries near to Russia present even a greater variety.
These problems present real challenges to plate recognition systems. MegaPixel's Systems
The first MegaPixel's system MegaCar was developed in 1996. The basis of the system was an ISA-bus Image Processor MegaFrame-3 from Megapixel. The frame size was 336*192*8 bit grayscale. The working range of the vertical size for standard size characters was 12-16 pixels (on the average 14 pixels). As the vertical size of the standard size characters is 80 mm (all European plates are the same), the field of view is 1.9 m:
The second MegaPixel's system CarFlow was developed in 1998. It was the PCI-bus Image Processor MegaFrame-4. The frame size for MegaFrame-4 is 768*288 gray or color pixels (rectangular, not square, pixels). MegaFrame-4 works much faster then MegaFrame-3. The first: PCI-bus much faster then ISA-bus. The second: MegaFrame-4 has the powerful hardware support of the Xilinx FPGA chip. This chip is suitable because the MegaPixel's neuro-like algorithms (Gaussian functions) use only 8 fixed bits while completely corresponding to the dynamic range of a real neuron - 1:10 Hence, there is no need for an expensive DSP solutions with floating point. Maximizing the resolution on the horizontal axis has allowed us to reduce the vertical size up to 10 pixels. It corresponds to a field of view approximately 3m: Width of a Field of View, Character's Resolution Today approximately 30 companies offer License Plate Recognition Systems including Automatic Number Plate Recognition and Electro-Optical Technologies, Inc. Unfortunately, the majority of the companies do not specify characteristics of the systems. Therefore, for comparative analysis, which is not complete, the systems of Talon from Racal Messenger Ltd., IMPC from Optasia Systems Pte Ltd. and materials from an excellent article "Opening the Shutters on Traffic Surveillance" from PULNiX America, Inc. were chosen. One of the requirements stated by Talon is "Size of a plate: Between 25% and 33% of horizontal field of view of camera". This corresponds to a field of view from 1.5m to 2.0m (length of all European plate is 0.5m) or a resolution from 15 to 20 pixels per character. One of the requirements stated by IMPS is "The character's height as captured would be at least 18 pixels high". This corresponds to a field of view 1.7m. All calculations are made for European license plates. The author of the Article writes, "For typical OCR reading, it is necessary to have at least 10 pixels/inch." This corresponds to a field of view 1.95m or a resolution 22.5 pixels per character (USA license plate, square pixel progressive scan camera is used). PULNiX America, Inc. is the World Leader in progressive scan technology. This company was the first to understand the necessity of creating special cameras for ITS (Intelligent Transportation System) applications. Progressive scan technology is necessary primarily to increase a field of view. PULNiX has developed two types of traffic cameras for ITS applications. One is the TM-9701TC, which is 768 x 480 pixels with a 4:3 aspect ratio and the TM-1001TC having 1K x 1K pixels with a 1:1 aspect ratio. PULNiX also creates the new 2K x 1K camera with 16 : 9 aspect ratio.
This picture is taken from article, of mentioned above. Note. In the article USA license plates are used which are of a different size.
This picture is taken from same article. The summary Table 1 contains results of calculations for various systems and cameras. Note: Russian license plates contain small character, which are 75% of standard. Therefore the table contains separate results for Russian license plates.
* - There are no data whether the double resolution on a horizontal is used. ** - These result are calculated from the assumption that at work with the standard camera (768x288) the double resolution on a horizontal is not used (384x288 is used only). Otherwise a field of view will be less. MegaPixel continually searches for other ways to increase a field of view. Today MegaPixel is testing the CarFlow II. In this system two synchronous standard cameras and one Image Processor MegaFrame-4 are used. The fields of view of the two cameras are overlapped. The resulting field of view of the system reaches 5.5m (for Russian license plates). The variants of parallel work two and more MegaFrame-4 for acceleration of work of system and increase of a field of view (at connection more then two cameras) are investigated. Speed of Processing One of the most important characteristics of a license-plate recognition system is the speed of processing. In MegaPixel's systems the original neuro-like algorithm of license plate detection is used. The algorithm works both for front and for rear license plates. The special external trigger is not necessary. This is very important for systems with a wide field of view, as there is often multiple vehicles in one image. The MegaPixel's algorithm is multizoned - up to 16 zones of license plates can be detected simultaneously. The speed of processing does not depend on number of the detected zones. After detection of a zone, the image of a license plate (192x24x8bit) is preprocessed. Removal of false detection, cleaning, definition of type - black or white, expansion and binarisation functions are performed. The final process is Optical Character Recognition (OCR). The processing time is approximately 20ms. All processing times are given for Pentium II/ 400 MHz processor. For comparison in system IMPS: "Response Time - time taken to read the plate number. This ranges from 0.4 to 2 seconds, configurable". Contrast of a License Plate Another important parameter of an LPR-system is minimal contrast of the image of a license plate. Talon defines "Contrast" as, "The difference between average character intensity and average plate intensity must be greater than 25% of difference between peak black and peak white luminance signals". Using this definition of contrast, MegaPixel's systems can work with 5%-contrast images of license plates. 5%-contrast license plate Maximal Speed of a Vehicle What is the maximal speed of a vehicle can be? This value depends from many conditions. The figure illustrates typical installation of a camera for LPR-system.
L - field of view on a horizontal (car flow direction); H - field of view on a vertical.
C - height of characters (meters); P - resolution (pixels per character); V - vertical resolution of a camera (for CCIR - 288 pixels, halfframe). ( V - P ) means, that license plate should be inside a field of view completely. The following is the resulting formula for calculation of maximum allowable vehicle speed:
Note: This formula is simplified and excludes some geometrical details. But under conditions: it works rather precisely. Table 2 contains results of calculations of maximal vehicle speed S for MegaPixel's systems (Russian license plates - 10 pixels per character) at various values of detection cycle time T and look down angel X. Table 2. MegaPixel's systems, maximal vehicle speed S (in Km/hour)
Table 3 contains results of calculations of maximal vehicle speed S for IMPS system (European license plates - 18 pixels per character) at various values of detection cycle time T and look down angel X. Note: the best T for IMPS is 400ms. The values 100ms and 200ms are taken from assumption, that the time of detection is less then complete cycle.
Thus it is possible to assume, that basically IMPS can work with external triggers only (for example, inductive loop sensors). Camera Installation The author of the Pulnix Article writes, "The typical look down angle is 40 degrees to avoid blocking the view by the following car". However, it is possible to object, following: Usually in MegaPixel's LPR-systems a lookdown angle from 20 up to 30 degrees are used (but large angles also work). Therefore the system has time to catch same license plate several times. The special algorithm uses it for increase of accuracy of recognition. The accuracy for Russian license plates is from 95% up to 98%. The Talon system also uses a smaller look-down angle: "Dip: Angle of camera boresight to horizontal should not exceed 20degrees It somewhat limits the system. Besides there are two limitations: "Skew: Angle of camera boresight to vehicle path should not exceed 20 degrees". Rotation: Must not exceed ±2degrees (unless rotation compensation enabled)". The same parameters for MegaPixel's system are: skew up to 45 degrees, rotation up to ±10 degrees. There is an exact dependence of a rotation-angle R on a dip-angles X and a skew-angle Y. It is geometrical distortions. But it is not a real rotation; it is a skew in a vertical direction.
Original image / Skew image (20 degree) Table 4 contains values a rotation-angle R as function from a dip-angles X and a skew-angle Y. Table 4. Rotation-angle R (degree)
For a stationary system these angles can be compensated for when the camera is installed. But for mobile systems, the new detection and compensation algorithms are necessary. Recently, MegaPixel developed a special mobile system for police cars in which the camera could rotate around a vertical axis and read license plates as they come in its field of view.
MegaPixel has been successful working with system integrators. In Russia the strategic partner of Megapixel is company-integrator JCS Rossi. For the System Integrator, MegaPixel offers the standard set: the Image Processor MegaFrame-4 board (boards) plus the DLL-software kernel (Borland C++). MegaPixel understands that a system integrator can make system interface better and faster (including use of OCR). MegaPixel stands ready to modify its software and the hardware to meet the business requirements of its customers. MegaPixel also welcomes joint projects.
Preprocessing time 3ms per zone OCR-time 20ms Dip-angles X 15-45 degree Skew-angle Y 0-45 degree Rotation-angle R up to ±10 degree Video Surveillance Motion Detector System MegaSense-8
MegaSense-8 is the MegaPixel's soft/hardware PC-compatible system intended to solve problems of security, supervision, registration and control of access. The system is implemented with the use of the latest neuro-like algorithms of digital image processing and high-speed hardware developed by MegaPixel. MegaSense-8 1. High-selective detection of movement 2. Wide capabilities of registrating videodata: 3. Enhanced means of visualization and processing of videodata: 4. Wide capabilities of generating spoken messages and controlling external executive devices: The system provides the following specifications: The main feature of this system is the use of neuro-like algorithms. The system functions without regard for changes of average brightness or contrast. I it maintains positions of object contours that allows it to avoid false alarms. For example, the system copes with two-multiple jumps of brightness, which can occur at work with luminescent illumination, at the value of a camera shutter - 1/500. While the MegaSense system was developed as security system, the principles incorporated in it can be used for creation ITS-systems: red-light running, speeding, rail crossing violations, toll violations, antiterror application, incident detection etc. MegaPixel is committed to assisting it partners and customers adapt this new technology to meet their business needs.
PoltechCORPORATE PROFILE
Product(s): License Plate Reader and Automatic Container Code Reader Description: The real-time reading capabilities of the Perceptics License Plate and Container Code Readers provide automatic vehicle and container code identification in multiple applications: Perceptics' LPR is comprised of a state-of-the-art imaging system and innovative decision-tree technology, and can be integrated into existing systems to expand their accuracy and capabilities or linked to databases for immediate access of current file information. An on-board operating system allows the LPR to operate in a stand-alone or background mode, while our Intelligent Character Recognition (ICR) guarantees correct reads despite variations in font, syntax, size, perspective and tilt.
|